Crash Risk and Market Resilience in Retail-Dominated Markets
Evidence from China’s 2024 Securities-Relending Suspension
Álvaro Cartea & Hao Ding  •  Oxford-Man Institute of Quantitative Finance, University of Oxford  •  Draft June 2026

Research Question

When a short-selling constraint binds completely in a retail-dominated market, does crash risk rise or fall? In the data it fell — and which flow replaces the short sellers decides the sign.

On 11 July 2024, China suspended securities relending — the wholesale supply of borrowable shares — producing a de facto elimination of short-selling capacity for most A-shares in a clean, non-crisis setting. Theory is split on the consequence for crash risk: removing short sellers could trap negative information and destabilise prices (the textbook bad-news-hoarding view), or, in a market where individuals are over 80% of volume, retail “buy-the-dip” liquidity could fill the vacuum and stabilise them. Which force dominates is an empirical question that earlier short-selling shocks — partial, temporary, and crisis-driven — could not answer cleanly.

Hypotheses & verdicts
H1 (policy efficacy) the ban cuts short interest in previously heavily-shorted stocks — confirmed.
H2 (crash risk) H2a information friction raises it  vs H2b liquidity provision lowers it H2b supported, H2a rejected; net crash risk decreased ↓.
H3 (investor heterogeneity) the effect varies by investor type — retail-herd stabilises, hot-money destabilises — confirmed.

Policy Timeline & Headline Effects

PRE-BAN Treat × Post = 0 POST-BAN Treat × Post = 1 — effects measured here 2020Q1 2025 margin hike 2023-10-30 T+0 crackdown 2024-02-06 RELENDING SUSPENDED 2024-07-11 · structural break legacy settlement 2024-09-30 HFT rules codified 2025-07-07 POST-BAN EFFECTS · Treat × Post POLICY BITE −31.5 basis points · short interest t = −21.1 · within-R² = 15.1% MAIN RESULT −0.073 crash risk (NCSKEW · IV) p < 0.001 · contra bad-news-hoarding MECHANISM · DTB DDD Retail herd ↓ −0.015 DUVOL · p=0.050 Hot money ↑ +0.013 p=0.009 ✓Holm retail stabilises · hot money destabilises · net ↓

The shock at a glance. Policy timeline around the 11 July 2024 relending suspension (the structural break) and the three headline post-ban effects on previously heavily-shorted stocks: short interest fell, crash risk fell, and the sign splits by investor type.

Key Results

The suspension sharply cut short interest in the most heavily-shorted stocks (−31.5 bp), yet their future crash risk fell rather than rose — the opposite of the bad-news-hoarding prediction. The sign is set by who replaces the short sellers: stabilising retail buy-the-dip flow dominates destabilising hot-money flow, and the retail share of trading volume rises in treated stocks after the ban. The one cost is everyday liquidity — treated stocks turn more illiquid — but wider price impact does not translate into more crashes.

FindingEstimateContext
Policy bit — daily short interest (H1)−31.5 bp***Most- vs least heavily-shorted (pre-ban), daily; within-R² = 15% (t = −21)
Future crash risk fell, IV / 2SLS (H2)−0.073***NCSKEW Treat×Post, p < 0.001; OLS −0.022, insignificant — contra bad-news-hoarding
Aggregate attention channel≈ 0Attention×Treat×Post insignificant in OLS & IV — not the channel
Retail-herd flow → crash risk (H3)−0.015*Stabilising (p ≈ 0.05); Dragon-Tiger Board branch volume
Hot-money flow → crash risk (H3)+0.013***Destabilising (p = 0.009, DUVOL); only channel to survive Holm–Bonferroni
Retail volume share (level-2, < ¥40k)+1.2 pp***Treated post-ban (t = 3.12, p = 0.002) — retail fills the vacuum
Off-market block trades−1.08***Treated vs control (p = 0.0015) — no substitution off-market
Index-futures migrationpartialVolume rose but open interest stayed flat-to-down — not a build-up of short positions
Illiquidity (Amihud) — the cost+4.97***Treated more illiquid post-ban (t = 5.91) — wider price impact

* p < 0.10;  *** p < 0.01. SEs clustered by firm. Parallel trends hold in the triple-difference. Red = destabilising / cost.

Mechanism

Dual-role mechanism

Removing short sellers cuts two ways: an information-friction channel (destabilising) and a liquidity-provision channel as retail buy-the-dip fills the vacuum (stabilising). In a >80%-retail market the latter dominates on net, so crash risk falls. Two independent lenses agree — Dragon-Tiger Board branch flow and market-wide level-2 composition both show retail taking up a larger share in treated stocks.

Why the pooled effect is muted

The stabilising retail-herd and destabilising hot-money channels carry opposite signs, so they partially offset when pooled into a single attention measure — hence the imprecise aggregate triple-difference. The IV direct effect is read as support that crash risk fell, not the causal magnitude: a complier (LATE) estimate with an imperfect exclusion restriction.

The cost, honestly stated

Treated stocks became more illiquid (Amihud +4.97) — wider per-trade price impact. But wider spreads are not more crashes: a steady stream of retail orders absorbs gradual selling pressure (lowering tail risk) even as per-trade impact rises — the level-2 retail-share rise is the direct evidence.

Data & Empirical Design

Sample & outcomes
Identification

Contribution & Implication

1.The crash-risk effect of removing — not introducing — short-selling capacity

Prior China studies show introducing short selling cut crash risk (Chang et al. 2007; Li et al. 2024); the crisis-era ban literature studied removal but focused on liquidity and price discovery in partial, crisis-driven bans (Beber & Pagano 2013; Boehmer et al. 2013). To our knowledge this is the first evidence on the crash-risk consequences of a clean, non-crisis, near-complete removal in a retail-dominated market: removal need not symmetrically raise crash risk when the post-suspension equilibrium is dominated by retail flows — directly against the bad-news-hoarding prediction (Jin & Myers 2006; Hutton et al. 2009; Hong & Stein 2003).

2.A composition-of-liquidity channel — who fills the vacuum

The crash-risk consequence is not a fixed property of the ban but of the marginal liquidity provider. Branch-level and market-wide level-2 data agree that retail flow lowers crash risk while hot-money flow raises it — adding investor-type heterogeneity to the short-sale-constraint and attention literatures (Miller 1977; Diamond & Verrecchia 1987; Barber & Odean 2008; Da et al. 2011).

3.Implication — for retail-dominated markets

The headline lesson is that a short-selling ban need not raise crash risk: the consequence hinges on the composition of replacement liquidity, so where retail buy-the-dip is deep the ban can calm tail risk — at the cost of worse everyday liquidity. Because the 2024 shock is non-crisis (unlike China’s confounded 2015 episode; Beber & Pagano 2013; Boehmer et al. 2013) and the design — previously heavily-shorted vs lightly-shorted quartiles with board×quarter fixed effects — is portable, the lesson transfers to the many >80%-retail emerging markets weighing similar measures.

Selected References

Miller (1977) • Diamond & Verrecchia (1987) • Hong & Stein (2003) • Jin & Myers (2006) • Hutton, Marcus & Tehranian (2009) • Chen, Hong & Stein (2001) • Beber & Pagano (2013) • Boehmer, Jones & Zhang (2013) • Chang, Cheng & Yu (2007) • Barber & Odean (2008) • Da, Engelberg & Gao (2011).